RLadies #18: Frequentist & Bayesian random effects models in R: RStan tutorial

This month's guest speaker Amanda Koepke will present on random effects modeling in RStan:

A common problem in science, medicine, and other fields is the need to combine independent measurements (taken over multiple days, or at different labs, or using different measurement methods) into a consensus estimate in a way that recognizes explicitly the contributions that different sources of uncertainty (i.e., within-study uncertainties and between-study differences) make to the overall uncertainty. Many methods exist for this, and past studies have compared the performance of different estimation approaches, however little attention has been paid to Bayesian approaches. In this talk, I will briefly describe my work extending these estimator comparisons to Bayesian models and exploring which approaches are best for common data scenarios at NIST (https://www.nist.gov/). I will then give a short tutorial on how to implement this type of Bayesian analysis in R using RStan. The Hamiltonian Monte Carlo used in RStan provides a very efficient way to explore the posterior distribution and allows for the most natural definition of specialized probability distributions. However, it can be tricky to consistently obtain valid samples when trying to fit many simulated datasets with realistic simulation settings. I'll describe some of the issues I encountered and how to deal with some common RStan warnings.